Detecting pairwise correlations in high-density recordings: open science in action
Stephen J Eglen
Acknowledgements
Catherine Cutts, Tom Edinburgh, Gerrit Hilgen, Evelyne Sernagor.
BBSRC, EPSRC, Wellcome Trust, software Sustainibility Institute.
Correlation spike trains
Retinal waves: Wong et al (1993) (fig 8, 9)
How to interpret figures?
- three regions: [0,1], 1, [1, ∞]
- lower firing rates -> higher CI
What method should we use?
Phase 1: finding a short list
34 measures in literature + 1 from us => 35.
Six necessary properties:
- Symmetric
- Robust to variations in firing rate
- Robust to amount of data
- Bounded [-1, +1]
- Robust to variations in bin width (Δt)
- Anticorrelation: The measure should discriminate between no correlation and anticorrelation.
Dependence on firing rate
Phase 2: Desirable properties
Desirable properties:
- D1: Ignore periods when both neurons are inactive.
- D2: minimal assumptions on structure.
- D3: aside from Δt, minimise number of parameters
Four methods:
- Kerschensteiner and Wong correlation (D1, D2)
- Spike count correlation (D2, D3)
- Kruskal et al. binless correlation measure (D1?, D2, D3)
- Tiling coefficient (D1, D2, D3)
Tiling measure
Evidence
Blankenship et al (2011)
Correlations decay with age
Inference methods
Biorxiv paper.
Correlogram methods
Extending across time with tau.
Looking for needles in haystacks.
Competitions: burst analysis
Ranking of burst analysis methods
Burst analysis: mouse retinal neurons
Burst analysis: networks dervied from human stem-cells
Reproducible research
To do this work we needed access to data. We have released this (Eglen et al 2014, Gigascience).
Then and now
